1 July 1991 in Journal of Parallel and Distributed Computing
20 June 2023 in ACM Transactions on Parallel Computing
The current trend of performance growth in HPC systems is accompanied by a massive increase in energy consumption. In this article, we introduce GreenMD, an energy-efficient framework for heterogeneous systems for LU factorization utilizing multi-GPUs. LU factorization is a crucial kernel from the MAGMA library, which is highly optimized. Our aim is to apply DVFS to this application by leveraging slacks intelligently on both CPUs and multiple GPUs. To predict the slack times, accurate performance models are developed separately for both CPUs and GPUs based on the algorithmic knowledge and manufacturer’s specifications. Since DVFS does not reduce static energy consumption, we also develop undervolting techniques for both CPUs and GPUs. Reducing voltage below threshold values may give rise to errors; hence, we extract the minimum safe voltages (
V
safeMin
) for the CPUs and GPUs utilizing a low overhead profiling phase and apply them before execution. It is shown that GreenMD improves the CPU, GPU, and total energy about 59%, 21%, and 31%, respectively, while delivering similar performance to the state-of-the-art linear algebra MAGMA library.
12 December 2011 in IEEE/ACM Transactions on Networking
L7-filter is a significant deep packet inspection (DPI) extension to Netfilter in Linux's QoS framework. It classifies network traffic based on information hidden in the packet payload. Although the computationally intensive payload classification can be accelerated with multiple processors, the default OS scheduler is oblivious to both the software characteristics and the underlying multicore architecture. In this paper, we present a parallelized L7-filter algorithm and an efficient scheduler technique for multicore servers. Our multithreaded L7-filter algorithm can process the incoming packets on multiple servers boosting the throughput tremendously. Our scheduling algorithm is based on Highest Random Weight (HRW), which maintains the connection locality for the incoming traffic, but only guarantees load balance at the connection level. We present an Adapted Highest Random Weight (AHRW) algorithm that enhances HRW by applying packet-level load balancing with an additional feedback vector corresponding to the queue length at each processor. We further introduce a Hierarchical AHRW (AHRW-tree) algorithm that considers characteristics of the multicore architecture such as cache and hardware topology by developing a hash tree architecture. The algorithm reduces the scheduling overhead to O(log N) instead of O( N) and produces a better balance between locality and load balancing. Results show that the AHRW-tree scheduler can improve the L7-filter throughput by about 50% on a Sun-Niagara-2-based server compared to a connection locality-based scheduler. Although extensively tested for L7-filter traces, our technique is applicable to many other packet processing applications, where connection locality and load balancing are important while executing on multiple processors. With these speedups and inherent software flexibility, our design and implementation provide a cost-effective alternative to the traffic monitoring and filtering ASICs.
1 September 2004 in Design Automation for Embedded Systems